Multi-brand routing on one line, surge predicted before it forms, per-brand P&L nightly, and aggregator commissions reconciled to the rupee — 16 delivery-only solutions, narrated by Iris.
A cloud kitchen is five brands, three aggregators and zero dining room — a business that exists only as order flow. When routing or commission reconciliation slips, no one sees it until the P&L does.
Rider shelf time — how long finished food waits before it leaves.
Commission — what each platform actually took, against what it should have.
Brand contribution — what a brand earns after commission, packaging and delivery.
Platform standing — rating, refunds and rank, which decide whether orders arrive at all.
No dining room, no walk-in, no second chance to fix an order in front of the guest. Everything that goes wrong goes wrong at a distance.
Multi-brand routing on a single production line — five brands sequenced as one queue, not five kitchens pretending to share a room.
Surge predicted rather than absorbed, because a delivery kitchen cannot ask the queue to wait.
Per-brand P&L nightly and commissions reconciled — so a busy brand is never mistaken for a profitable one.
Rider shelf time, commission, brand contribution and platform standing — live, with Iris naming the brand that is busy and unprofitable.
In a delivery-only kitchen the till is not a counter — it is the single point where five brands and three aggregators become one sequenced order book.
Five brands, one order book. Every aggregator and every brand writes into the same queue rather than a tablet farm on a shelf.
Every line knows its brand. Commission, packaging and delivery cost attach at the line, so brand contribution is computable from the first order.
Direct channel prioritised. Iris surfaces the first-party route where it wins, because a direct order and a marketplace order are not worth the same.
Rider shelf time, commission, brand contribution and platform standing on one live board — with Iris naming the brand that is busy and unprofitable.
Per brand, not per site. A dark kitchen running five brands needs five P&Ls, not one blended number that hides the loss-maker.
Per platform, too. The same dish sold on three aggregators earns three different margins, and the board shows all three.
Iris names the brand. Not that contribution fell — which brand, on which platform, and whether it is volume, commission or packaging.
P&L, balance sheet, cash and close on a single balanced ledger — with every aggregator payout reconciled against what the orders said it should be.
Payouts checked, not accepted. Each platform remittance is matched line by line against the orders behind it, and the gap is named.
Revenue split by brand automatically. One kitchen's takings resolve into per-brand revenue without a spreadsheet at month-end.
Ask Iris for any number. Any brand, any platform, any period — answered in the conversation, not a week after close.
Demand-planned replenishment for a dark store feeding several brands from shared stock — where one ingredient shortage can take four menus offline at once.
Shared stock, several menus. Ordering accounts for every brand drawing on the same ingredient, not each brand forecasting alone.
Shortage impact is visible upstream. Before a line runs out, Iris shows which brands and how many listings it will take down.
Feeds the FIFO ledger. Receipts post straight into dark-store depletion, so rotation is enforced rather than intended.
Digital checklists, shift routines and standard operating procedures across every kitchen — completion tracked live, exceptions escalated.
Assembly checks, signed off. The pack-out routine that decides whether an order arrives complete is recorded, not assumed.
Shelf time tracked per order. How long finished food waits for a rider rolls up by brand, hour and platform.
Exceptions escalate themselves. The routine that didn't complete surfaces to a human; the rest stays quiet.
Rostering, attendance and labour percentage measured against forecast demand. Iris flags the overstaffed lulls and the understaffed peaks before the shift, not after.
Rostered to predicted surge. Kitchen cover is built from the surge forecast, which in delivery-only is the only demand signal there is.
Labour measured per order, not per cover. With no dining room, orders per kitchen hour is the productivity number that matters.
Flagged before the shift. An understaffed peak is caught while the roster can still change — and in this sector a missed peak is a rank penalty.
Deterministic logic under the hood (the “Zen Rules” layer) — stock rotation and system connections behave identically every time. This sector is unusually light here: only two of sixteen solutions are deterministic. Almost everything else is a model.
One platform's menu API is failing on write. Two brands are at risk of selling an item the dark store no longer has. Raised before a refund, not after one.
The deterministic half of dark-store stock — depletion through the recipe, first-in-first-out rotation enforced, and a sell-out that removes an item from every brand and platform in the same second.
Every order depletes shared stock through its recipe, so inventory is a consequence of trading rather than a nightly count.
FIFO rotation is enforced at pick, not suggested — in a dark store nobody is looking at the shelf to catch it.
One ingredient is shared across brands, so a sell-out pulls every affected listing on every platform simultaneously.
Batch and expiry travel with the stock, with alerts raised before product has to be written off.
Transfers between kitchens keep lot identity, so traceability survives stock moving between sites.
In delivery-only, integrations are not plumbing — they are the business. Payments, every aggregator, tax per market and the accounting ledger, connected and health-checked continuously.
Every aggregator connects through one hub, so adding a platform is configuration rather than a project.
Connection health is checked continuously with last-sync times and error rates — a degraded API is caught before it costs orders.
When a platform does fail, Iris quantifies the orders and the brands at risk rather than raising a generic error.
Tax rules are configured per market, which is why launching a new city or country takes days rather than a release.
Sales, commission, refunds and settlement post themselves to the accounting system, already coded to the right accounts.
Forecasting, optimisation, attribution and pricing models under the hood (the “Zen Models” layer) — where a business made of pure order flow actually lives. Fourteen of the sixteen solutions run here.
Five brands, three aggregators, one production line. Orders are sequenced across the whole kitchen by promise time and station load rather than by which brand shouted first.
Every brand and platform writes into one sequenced queue, so the line works a single list rather than five competing ones.
The optimiser sequences by promise time and station load, batching shared prep steps across brands where they overlap.
Brand identity survives the merge — packaging, label and insert stay correct even when two brands share a fryer.
Re-sequencing happens automatically when a station backs up, rather than a manager reshuffling tickets by hand.
Capacity is protected per brand, so one brand's promotion cannot starve the other four of line time.
The moment that decides whether food arrives hot and whether it arrives at all. Handover is gated, timed and matched — because a wrong bag given to a rider is unrecoverable.
Order and rider are matched at the gate by scan, so the right bag leaves with the right person every time.
Shelf time is measured from ready to collected, which is the interval that quietly ruins the food and the rating.
Production is paced against predicted rider arrival, so food is finished as the rider lands rather than long before.
A late or no-show rider raises early enough to hold production instead of cooking into an empty shelf.
Handover is timestamped and evidenced, which settles a “never arrived” dispute with a record rather than an argument.
The short-horizon model — the next thirty to ninety minutes. A delivery kitchen cannot ask a queue to wait, so the surge has to be seen while there is still time to fire prep.
Order arrival is predicted in short windows, giving the line enough warning to start prep rather than react to a backlog.
Weather, local events and platform promotions are read as surge triggers rather than explained afterwards.
Predicted surge feeds promise times directly, so quoted delivery windows stay honest when volume climbs.
When capacity will genuinely be exceeded, Iris recommends throttling a platform before ratings take the damage.
Surge by brand is separated, since a promotion on one brand does not lift the other four.
The long-horizon model — days and weeks ahead. What to prep, how much to order and how many people to roster, forecast per component across every brand drawing on it.
Time-series forecasting projects demand by day and daypart per brand, then aggregates to the shared component level.
Prep quantities are set against forecast rather than against what was prepped last week.
Component shelf life bounds the batch, so the plan never prescribes prepping more than can sell before it expires.
The same forecast drives ordering and the roster, so stock, labour and plan cannot contradict each other.
Forecast error is tracked per brand and fed back, so a consistently over-forecast brand corrects itself.
The forecasting view of dark-store stock — projected cover per component, expiry risk ahead of time, and the replenishment call made before a shortage takes listings down.
Stock cover is projected against forecast demand, so a shortfall is a date on a calendar rather than a surprise at 8pm.
Expiry risk is forecast per batch, flagging what will not sell in time while it can still be promoted or rerouted.
Shared components are projected across every brand consuming them, not brand by brand in isolation.
Reorder points learn from lead time and demand volatility rather than sitting at a number set on launch day.
The listings a shortage would take offline are named in advance, so the commercial cost of a stockout is known before it happens.
In delivery-only, packaging is a real cost line and a missing item is a refund. Assembly is verified and packaging is costed into the order it belongs to.
Every order carries a checklist at pack-out, so a missing side is caught in the kitchen rather than at the customer's door.
Packaging is costed into the line — boxes, bags, sleeves, cutlery, seals — so margin per order is real rather than food-cost only.
Anomaly detection links refund and complaint patterns back to specific assembly steps, brands and shifts.
Packaging spec is matched to journey time, so a long delivery is not packed like a short one.
Packaging consumption depletes stock like any other input, so a run-out is forecast rather than discovered mid-rush.
Menu, price and availability held as one source of truth across every platform and every brand — so nothing drifts, and nobody can order what the dark store no longer has.
One menu feed powers every aggregator for every brand — no per-platform spreadsheet, no re-uploads.
A price or availability change lands everywhere at once, so the same dish cannot be two prices on two platforms by accident.
A sell-out pulls the item from every affected brand and platform in the same second.
Store open, pause and prep-time settings are managed centrally rather than in each platform's own dashboard.
Sync failures are surfaced with the orders at risk quantified, rather than discovered when a refund arrives.
Commission is the largest controllable line in this sector and the least examined. Every rate is modelled, every payout is checked, and platform pricing is set deliberately rather than uniformly.
Effective commission is computed per order — base rate, promotion share, delivery fee and payment charge together, not the headline percentage.
Platform pricing is set to protect net margin after commission, so a marketplace order is not quietly sold at a loss.
Parity rules are enforced where a platform requires them and deliberately broken where it does not, rather than defaulting to identical everywhere.
Each remittance is reconciled line by line, and any gap between what was charged and what was contracted is named.
Promotion participation is measured on incremental orders rather than gross uplift, which is the only honest read.
One kitchen's takings resolved into per-brand revenue automatically — and every platform payout matched against the orders that should have produced it.
Revenue attributes to brand at the line, so a shared kitchen never needs a month-end spreadsheet to split its takings.
Each platform remittance is matched to the orders behind it, and unmatched or short-paid orders are listed rather than absorbed.
Refunds and chargebacks settle against the brand that incurred them rather than against the site as a whole.
Settlement timing differences between platforms are modelled, so cash position is accurate rather than optimistic.
Franchise or brand-partner splits are computed automatically where a kitchen operates someone else's brand.
A nightly P&L per brand, per platform — because in a five-brand kitchen the blended number is the one that hides the brand losing money on every order.
Food, packaging, commission, promotion and delivery are attributed per brand, giving true contribution rather than revenue.
Shared costs — rent, utilities, core labour — are allocated on a defensible driver rather than split evenly.
The same brand is compared across platforms, since a brand can be profitable on one and loss-making on another.
Contribution per kitchen hour is computed, which is the number that decides which brand deserves the line at peak.
The P&L runs nightly rather than monthly, so a brand that turned unprofitable is caught in days.
Which brands earn their line time, which are busy and unprofitable, and which should be retired — ranked on contribution per kitchen hour rather than on order count.
Brands are ranked on contribution per kitchen hour, which is the scarce resource a shared kitchen actually sells.
Attribution separates brand performance from platform effects, so a rank drop is not mistaken for a demand drop.
Cannibalisation between your own brands is detected, since two similar concepts on one platform often split one demand pool.
Iris flags the busy-but-unprofitable brand explicitly, because volume is the easiest thing to mistake for success here.
Retire, reposition or reprice recommendations come with the modelled effect on the rest of the portfolio.
Running the portfolio — which concepts to hold, where the demand gaps are in your own catchment, and whether a new brand would take share from a competitor or from you.
A market gap model reads unmet demand by cuisine, price band and daypart in the delivery radius you actually serve.
Gaps are scored against your existing kitchen capability, so a concept you cannot produce well is not recommended.
Self-cannibalisation is modelled before launch, not discovered after two brands split the same orders.
The portfolio is managed as a whole — how many brands one line can carry before sequencing quality degrades.
Brand identity, menu and positioning are held centrally, so a concept can run consistently across several kitchens.
The mechanics of standing a brand up. A new concept is cloned, priced, published to every platform and live in days — because in delivery-only, launch speed is the whole competitive advantage.
A proven menu is cloned to a new brand or a new kitchen with its recipes, costs and prep steps intact.
Platform-specific pricing and listings are generated for every aggregator at once rather than built by hand three times.
Launch demand is forecast up front, so the first week is staffed and stocked rather than survived.
The new brand inherits the shared component plan, so it draws on existing stock instead of a separate supply chain.
Performance is measured against the launch forecast from day one, so a concept that is not working is retired quickly.
On an aggregator, your rank decides whether orders arrive at all. Ratings, refunds and standing are treated as an operational metric with a cause, not as feedback to read later.
Ratings and reviews are pulled from every platform into one view per brand rather than checked in three dashboards.
Sentiment is themed automatically — late, cold, missing item, portion, packaging — so a pattern surfaces without reading every review.
Refunds are traced to their operational cause and to the brand, shift and assembly step that produced them.
Standing is tracked as a leading indicator, since a rank penalty costs order volume long before it costs a rating point.
Recovery is triggered while it still matters, and repeated refund abuse is flagged rather than paid indefinitely.
Not a screenshot — the actual agent. No solution in this sector is filed under L3, because none needs to be: Iris queries all sixteen, across every brand and every platform, in plain language.
Sixteen solutions. Every card opens a live guided demo.
You configure a sector playbook, not a custom project. Iris applies it herself — agentically, from day one.
Onboarding is agentic. Iris connects every aggregator, brand and payment feed directly — no manual data mapping.
A new brand is configuration. Standing up a concept is a clone and a publish, not a project — which is what makes a portfolio strategy possible at all.
A new market takes days. Tax and platform rules are configured per market rather than coded per customer.
The first line of support is agentic — Iris resolves most of it herself. Our engineers pick up from there.
Layer 1 — Iris, 24/7. Configuration questions, anomalies and routine issues resolved directly, instantly.
Layer 2 — our engineers. Anything Iris can't close escalates automatically to a Zentallio engineer.
No blank tickets. Every escalation arrives with Iris's own diagnosis — engineers start from an answer.
“A cloud kitchen is five brands, three aggregators and zero dining room — a business that exists only as order flow. When routing or commission reconciliation slips, no one sees it until the P&L does.”
Zentallio is built for the invisible restaurant: multi-brand routing on one line, surge predicted before it forms, per-brand P&L nightly, and aggregator commissions reconciled to the rupee. Underneath sits the same platform every sector runs — one data spine from the till to the ledger, three layers of intelligence, and an agent that narrates every screen, flags what needs a human, and acts on the rest.
A brand that is busy and unprofitable, hidden inside one blended kitchen P&L.
Commission taken at a rate nobody reconciled, on every order, for months.
Food finished early, sitting on a shelf, arriving cold — and the rank penalty that follows.
Stock depletes and rotates deterministically, and every platform connection is health-checked continuously.
Routing, surge, commission and per-brand contribution are modelled — fourteen of the sixteen solutions live here.
“Which brand is busy and unprofitable?” — answered across every brand and platform in one question.
Every Zentallio sector runs the same three-layer intelligence, tuned to how that format actually loses margin.